categories

archives

meta


"walls of the city" logo conceptualized by Oleg Volk and executed by Linoge. Logo is © "walls of the city".

correlation does not imply causation

Correlation does not imply causation.

That is a sentence which – in various phrasings and rewordings – I have repeated at least twice a year for the past three years in my Graphics Matter series of posts. But, really, for the non-mathematicians amongst us (including myself), what does that actually mean?

Well, by way of the inimitable genius behind Dilbert, we have this example:

dilbertcorrelation

And today, by way of Bloomberg Businessweek, we have this representation:

businessweekcorrelation

Proving causality is, to put it very simply, a pain in the ass, and does not simply consist of "X happened, then Y happened, which means X caused Y". This is exactly why I do not claim any causality in my Graphics Matter series, and exactly why disproving the hypotheses of "guns cause deaths" and "guns cause crime" is so easy.

Remember that the lie may not be in explicitly what they say, but rather their underlying reasons for saying so – on the surface, the hypothesis that Dilbert was sending the link seemed valid, but the boss’ reasoning was fundamentally flawed. In his case, it was not a lie simply because he did not know any better, but in the case of those who would forcibly deprive us of our Constitutionally-protected individual rights, the same cannot be said.

(Dilbert strip brought to my attention by Barron Barnett – guess I need to read my "Funnies" RSS feeds before I read my serious ones… Second graphic courtesy of Say Uncle.)

4 comments to correlation does not imply causation

  • I’ve always been a fan of the Latin: “Post hoc, ergo propter hoc.”

  • BalloonGoesUp

    Just as important is to be able to correctly identify the dependent variable.

    In a recent post I was accused of rigging data to “prove” that my carry gun was the best based on my requirements. It didn’t strike them that my carry gun may be my carry gun BECAUSE it is the best based on my requirements.

  • I always liked this XKCD comic on the subject:

    http://xkcd.com/552/

  • @ Erin Palette: While the Latin is more concise, I would be willing to wager more people might understand the English faster ;).

    @ BalloonGoesUp: I always preferred identifying the independent variable first, but that is simply a matter of taste, and you get to the same end result regardless. It is the mushy stuff in between the dependent and independent variables where people get confused (for example, the idiotic notion that “more guns = more crime”).

    As for the post… well, it is always amusing when people get their causality chains confused… ;)

    @ Axess Denyd: Bwah! Sounds about right :).